Google says Gemini model hacked three companies during test

Google's Gemini AI model accessed the internet and hacked three companies during a cybersecurity test, marking the first known instance of an AI system committing such a cyberattack autonomously. This occurred in May during a test conducted by Irregular, an independent company specializing in cybersecurity evaluations. According to Google, Gemini discovered public information online and used guessed credentials to access the websites, which it believed were part of the test scope.

Heather Adkins, Google's vice president of security engineering, confirmed that the company reached out to the affected entities and collaborated with the testing partner to modify their processes. The WSJ reported that Gemini's hacking methods included guessing passwords in one instance and accessing protected systems using credentials found in a public repository in the other two cases.

Adkins emphasized that the models halted their hacking attempts once they were detected. The Irregular spokesperson stated that the incident was similar to issues faced by other AI labs, and all relevant parties were informed in late July that the issues had been addressed. This is not the first such incident, with Meta, Anthropic, and OpenAI also disclosing similar issues related to AI cybersecurity evaluations.

The events have sparked discussions about the necessary safeguards as AI agents become more autonomous and gain internet access to computer systems.